System Dynamics: Taming Expert Systems In the Business World
Mihran Markarian
Roman Koziol
37 Penkivil Street, Willoughy NSW 2068
Australia
Abstract
The paper reports on a new approach for the building of Decision Support Systems based on
System Dynamics and Expert Systems. The power of this approach is illustrated by using it to
identify problems that exist within the production processes of a manufacturing company. System
Dynamics was used to simulate the production processes and build.the expert system. The
simulation identifies the process where the problem exists, and the Expert System suggests
possible causes to the problem and the solutions required to bring production back to normal.
Stella was the System Dynamics tool used to gain a detailed understanding of the production
processes and their interactions in order to build a simulation model. The influences from this
simulation model were used to structure the knowledge base of an expert system. While the expert
system was an essential ingredient of the Decision Support System, the actual system that was
used and its features were of secondary importance.
This paper will benefit System Dynamic practitioners who are interested in:
1) simulating a process within an organisation.
2) the application of System Dynamics to solve a manufacturing problem.
8) _ the relationship between System Dynamics and building of Expert Systems.
© Copyright 1991
Performance Management Systems Pty Lid
Page 336
. System Dynamics '91 Page 337
4) _ the use of System Dynamics as a decision making/support tool in the manufacturing
industry.
Introduction
The manufacturing industry has become a vastly complex world that is both unpredictable and
difficult to manage. Control of production processes in this environment has become increasingly
difficult. Management decisions have largely been based on "gut feeling" and "educated guesses".
These traditional techniques have been largely encouraged by the lack of accurate information on
the current situation within the production processes, and managements lack of detailed
understanding of the manufacturing processes themselves. The decision support tools available are
inadequate for identifying the problem areas, thereby impeding the ability to focus on solving the
real problem(s).
The-work in this paper demonstrates the capabilities of a Decision Support System to identify
problems that exist in the production operation of a pacemaker manufacturing company. In this
case, the most common problems included delays, process bottlenecks, material shortages and
resource constraints.
Manufacturing Operations
The company is an international organisation specialising in the design and manufacture of human
implantable medical devices. The company has manufacturing operations in the Asia/Pacific, North
America and European regions. Research and Development is based in Sydney Australia. The
© Copyright 1991
Performanos Management Systems Ply Ltd
Page 338 System Dynamics '91
company currently ranks number three position in the world market place.
The Manufacturing Plant
The plant has the capacity to produce in excess of a thousand devices per month. Because each
device is for human implant, production standards are much more stringent than those normally
found in the electronics industry. "Good Manufacturing Practices" are employed to satisfy regulatory
authorities and ensure that a safe and reliable product is being produced. The. devices pass through
several cleaning and testing stages to ensure a contaminant free, high quality product.
Decision Support Systems In Manufacturing
There are many implementations of Decision Support Systems in industry today, and each one is
unique in its own right. There are Executive Information Systems, Simulation Systems and Expert
Systems, to name just a few, that attempt to provide the basis for more informed decision making.
Executive Information Systems translate real data, obtained from organisational operations, toa
graphical form for analysis. The result, is merely a snapshot of a constantly changing situation. The
picture conveyed, based on static data, seldom provides a suitable basis for making meaningful
projections into the future. This is because the projection has been made on the basis of static and
historical data, rather than the dynamic forces at work.
Simulation has been widely accepted by the manufacturing industry for optimising existing
operations. Simulation allows the application of what-if scenarios to visualise and assess the impact
of possible alternatives. However unforseen events that subsequently arise during normal operation,
© Copyright 1991
Performance Management Systems Pty Lid
System Dynamics '91 Page 339
that were not included in the scenario analysis, are resolved by the traditional methods mentioned
earlier.
Current implementations of Expert Systems rely on the accuracy of the input provided by the
experts in the field to.build a knowledge base that can be directed towards solving problems. Expert
Systems generally undergo many iterations of testing until the results demonstrate a high degree of
correlation between the recommendations of the expert system and that of the experts. This
process can be indefinite and the knowledge base may contain unnecessary, redundant or
insufficient knowledge. The problem is that there is no precise way of building the exact knowledge
base that is required to solve the problem.
This paper postulates an approach for building an exact knowledge base that is required to
address any problem.
Performance Decision System
The Decision Support System built was based upon using System Dynamics to obtain a sound
understanding of the processes in which a problem could occur. A Stella simulation was then used
to determine the location and the magnitude of the problems. Once the processes were understood
and the simulation was validated, the causes of the problem(s) became apparent and could be
captured in a knowledge base of possible solutions. This knowledge base was then used to
recommend the appropriate ‘solution when the same or related problems occurred in the future.
© Copyright 1991
Performance Management Systems Ply Lid
Page 340 System Dynamics '91
Why Stella
Stella is a computer software implementation of System Dynamics principle that can be used to
simulate processes. Stella is a powerful computer tool that can be used to model organisational
behaviour, by capturing the influences that can lead to the disruption of a steady state system.
Stella was found to be suitable for achieving understanding of the production processes. The ease
of use and the graphical interface allows the user:to focus on modelling the problem rather than
mastering the mechanics of the program.
Model of the Manufacturing Plant
The simulation comprised of a Main Model and several Sub-models. The Main Model
(refer Figure 1) is a high level view that represents the manufacturing operation with sufficient detail
to identify the main processes. The Main Model explodes out to the sub-models that detail the
influences affecting each process.
The sub-model includes the resources that are required by each process. The effectiveness with
which these resources are balanced determines the performance of the manufacturing operation.
The sub-models were developed with the aid of an expert who was responsible for the
manufacturing operation. The aim of the sub-models was to duplicate the same rules that the expert
used to allocate resources towards achievement of daily production targets (refer Figure 2). The
manufacturing plant was simulated using typical plant disturbance data to establish that the model
was capable of giving a true reflection of plant performance (refer figure 3).
© Copyright 1991
Performance Management Systems Pty Lid
System Dynamics '91 Page 341
The Expert System
When an expert system is being built, there is seldom an understanding of the processes and the
decisions that are involved. So the classical. approach is to elicit this knowledge from an expert and
to reproduce the same knowledge in the knowledge base of an expert system.
This knowledge base is often found to be made of adhoc rules and rule hierarchies. The adhoc
rules can represent unnecessary and redundant knowledge, whereas the rule hierarchies represent
the cause-effect relationships. It is a Systems Dynamics model which can provide the framework for
the discovery of these rule hierarchies. After all a System Dynamics model is nothing more than a
hierarchy of influences and cause-effect relationships.
It is the mapping of the hierarchy of influences to the hierarchy of expert rules that provided the
foundation upon which the Performance Decision System was built.
Conclusion
Based on the work carried out and the results obtained, it was demonstrated that a System
Dynamics approach can be used to:
eunderstand and simulate manufacturing processes
«solve a manufacturing problem
sestablish the relationship between System Dynamics and building of Expert Systems
ebuild an expert system based decision support tool.
The most significant aspect of this work was establishment of the equivalence between the
hierarchy of influences in a System Dynamics model to the rule hierarchy contained in the
knowledge base of an expert system.
© copyright 1991
Performance Management Systems Pty Lid
Figure 1. The Main Model
© Copyright 1991
Performance Management Systems Pty Ltd
Ce weg
16, sormeudg urayshg¢
test_o9_10_ WAT
Mad
tst_eq_avail final
Figure 2. The Sub-Model
toat_og.s09_WAT tat_og_avall WAT
fon
test_oq_roq_Jot total om ogulp
text_oq_roq_batt
test_oq_roq it text_oq_avail_jnit
{otal_tost equip
od
tost_oq_req_batt tost_oq_nvail iit
toxt_og_roq.fint
tost_oa_roq_tint sect og.avalLiinal
test_oq_soq_init fotal_test_oquip
Final Test 1961_eq_prd_atfiot ” j6s1_oq_yoq batt {st_og_unavail fat
tost_oq_soa_WAT
© copyright 1991
Performance Management Systems Pty Ltd
16, sormeudg wayskg¢
eve o8eq
Page 344 System Dynamics '91
Figure 3. EFFECT OF TEST EQUIPMENT
1 WAT 2 Finished Goods
60.90.
2 18888
4599 4
2 73.00
2 2
=
5 3000 |
22 3888
i=} > 2
= + 1 1 1 1 1:
is09 4
2 23.00
2 88 1.00 , 25.75 . 50.50 , 75.25 , 100.00
Time (days)
Testing Equipment
out of
1 WAT 2 Finished Goods Commission
for 1 week
60.90 4 ss
2 100.00
Effect of Equipment
Downtime
45.00 4 L
2 75.00
3 2
S 3000 | L
2 5 50.00 2
° n °
= 1 1 1 1
15.00 4
2 25.00
> 90 : — : :
1.00 25.75 50.50 75.25 100.00
Time (days)
© copyright 1991
Performance Management Systems Pty Ltd
System Dynamics '91 Page 345
References
Richomond, B., Vescuso, P., and Peterson, S. 1987. Business User's Guide to STELLA. Lyme, New
Hampshire: High performance Systems Inc.
© Copyright 1991
Performanoe Management Systems Pty Ltd
CC BY-NC-SA 4.0